4.1 The Riemann Integral
Darboux sums squeeze the area from above and below: Riemann's criterion, Riemann sums, the integrability of continuous and monotone functions, Lebesgue's criterion (discontinuities of measure zero) and both halves of the fundamental theorem of calculus.
What is the area under the graph of a function? For a polygon we can count triangles; for a curved region we must approximate, and the integral is whatever the approximations settle down to. Riemann (1854) made this precise with sums $\sum f(\xi_k)\,\Delta x_k$ over finer and finer partitions; Darboux (1875) streamlined it by trapping the area between an upper sum built from suprema and a lower sum built from infima. A function is integrable exactly when the trap can be closed. The surprising part is how much a function can misbehave and still be integrable — Thomae's function, discontinuous at every rational, has integral $0$ — and how little it takes to break integrability — the Dirichlet function, which only distinguishes rationals from irrationals, has no integral at all.
This section builds the Riemann integral from scratch, proves which functions are integrable, and ends with the fundamental theorem of calculus, which links integration to the derivatives of Chapter 3. CSIR NET tests it in three recurring ways: limits of sums that are Riemann sums in disguise (Part B), derivatives of integrals with variable limits (Part B), and "which of these functions are Riemann integrable / which statements about $F(x)=\int_a^x f$ are true" (Part C), where Lebesgue's criterion — bounded, with a set of discontinuities of measure zero — decides almost every option in one line.
Partitions, Darboux Sums and Riemann's Criterion
Throughout, $f:[a,b]\to\mathbb{R}$ is a bounded function, $a\lt b$. Boundedness is part of the definition of the Riemann integral: an unbounded function is never Riemann integrable in the proper sense (its integral, if any, is improper — §4.2).
A partition of $[a,b]$ is a finite set $P=\{x_0,x_1,\dots,x_n\}$ with $a=x_0\lt x_1\lt\cdots\lt x_n=b$. Its subintervals are $I_k=[x_{k-1},x_k]$, of lengths $\Delta x_k=x_k-x_{k-1}$, and its mesh (norm) is $\|P\|=\max_k\Delta x_k$. A partition $P^*$ is a refinement of $P$ if $P\subseteq P^*$. Any two partitions $P,Q$ have the common refinement $P\cup Q$.
The uniform partition $P_n$ has $x_k=a+k\,\frac{b-a}{n}$, so $\|P_n\|=\frac{b-a}{n}$.
For a partition $P$ put $M_k=\sup_{I_k}f$ and $m_k=\inf_{I_k}f$ (finite, since $f$ is bounded). The upper and lower Darboux sums are
$$U(f,P)=\sum_{k=1}^{n}M_k\,\Delta x_k,$$ $$L(f,P)=\sum_{k=1}^{n}m_k\,\Delta x_k.$$If $m\le f\le M$ on $[a,b]$, then $m(b-a)\le L(f,P)\le U(f,P)\le M(b-a)$. Geometrically, $U$ is the area of a staircase lying above the graph and $L$ the area of one lying below it.
(b) For any two partitions $P,Q$ of $[a,b]$: $L(f,P)\le U(f,Q)$. Every lower sum is at most every upper sum.
Proof. (a) By induction it is enough to add one point $y$ to $P$, say $x_{k-1}\lt y\lt x_k$. Only the term of $I_k$ changes: it is replaced by $\big(\sup_{[x_{k-1},y]}f\big)(y-x_{k-1})+\big(\sup_{[y,x_k]}f\big)(x_k-y)$. A supremum over a smaller set is no larger, so each of the two new suprema is $\le M_k$, and the new contribution is $\le M_k(y-x_{k-1})+M_k(x_k-y)=M_k\Delta x_k$. Hence $U(f,P^*)\le U(f,P)$. The lower sums are handled in the same way, with infima over smaller sets being no smaller; and $L(f,P^*)\le U(f,P^*)$ because $m_k\le M_k$ termwise.
(b) Apply (a) to the common refinement $P\cup Q$, which refines both: $$L(f,P)\le L(f,P\cup Q)\le U(f,P\cup Q)\le U(f,Q).$$ ∎
The upper and lower integrals of $f$ over $[a,b]$ are
$$\overline{\int_a^b}f=\inf_P U(f,P),\qquad \underline{\int_a^b}f=\sup_P L(f,P).$$By Theorem 4.1(b) the set of lower sums lies entirely below the set of upper sums, so both numbers exist (completeness, §1.2) and $\underline{\int}f\le\overline{\int}f$. We say $f$ is Riemann integrable on $[a,b]$, written $f\in\mathcal{R}[a,b]$, if $\underline{\int_a^b}f=\overline{\int_a^b}f$; the common value is $\int_a^b f=\int_a^b f(x)\,dx$. We set $\int_a^a f=0$ and $\int_b^a f=-\int_a^b f$.
Proof. For every partition $P$, $L(f,P)\le\underline{\int}f\le\overline{\int}f\le U(f,P)$. ($\Leftarrow$) If $U(f,P)-L(f,P)\lt\varepsilon$, then $0\le\overline{\int}f-\underline{\int}f\lt\varepsilon$; as $\varepsilon$ is arbitrary, the two integrals are equal.
($\Rightarrow$) Let $I=\int_a^bf$ and $\varepsilon\gt0$. By the $\varepsilon$-characterisation of $\inf$ and $\sup$ (§1.2) there are partitions $P_1,P_2$ with $U(f,P_1)\lt I+\varepsilon/2$ and $L(f,P_2)\gt I-\varepsilon/2$. For $P=P_1\cup P_2$, Theorem 4.1(a) gives $U(f,P)-L(f,P)\le U(f,P_1)-L(f,P_2)\lt\varepsilon$.
The last statement: $I$ and both sums lie in $[L(f,P_j),U(f,P_j)]$, an interval whose length tends to $0$. ∎
Riemann's criterion is the working tool: to prove integrability you do not need to know the value of the integral — you only need partitions on which the oscillation of $f$, weighted by length, is small.
Riemann Sums and Classes of Integrable Functions
Darboux sums need suprema and infima, which are often hard to compute. Riemann's original definition uses the value of $f$ at one chosen point of each subinterval instead.
Proof. ($\Rightarrow$) Let $|f|\le M$ (with $M\gt0$), $I=\int_a^bf$ and $\varepsilon\gt0$. Choose $P_0$ with $U(f,P_0)-L(f,P_0)\lt\varepsilon/2$, and let $m_0$ be the number of points of $P_0$ inside $(a,b)$. Put $\delta=\varepsilon/(8Mm_0+1)$ and let $P$ be any partition with $\|P\|\lt\delta$. Form $P\cup P_0$ by adding the $m_0$ points of $P_0$ to $P$ one at a time. Each added point splits one subinterval $J$ of length $\le\|P\|$, and, as in the proof of Theorem 4.1, changes the upper sum by at most $(\sup_Jf-\inf_Jf)|J|\le2M\delta$. Hence
$$U(f,P)\le U(f,P\cup P_0)+2Mm_0\delta\le U(f,P_0)+2Mm_0\delta,$$and similarly $L(f,P)\ge L(f,P_0)-2Mm_0\delta$. Subtracting, $U(f,P)-L(f,P)\lt\varepsilon/2+4Mm_0\delta\lt\varepsilon$. Both $I$ and every Riemann sum $S(f,P,\xi)$ lie in $[L(f,P),U(f,P)]$, so $|S(f,P,\xi)-I|\lt\varepsilon$.
($\Leftarrow$) Suppose $|S(f,P,\xi)-I|\lt\varepsilon$ whenever $\|P\|\lt\delta$. Fix such a $P$. Choosing each tag $\xi_k$ with $f(\xi_k)\gt M_k-\varepsilon/(b-a)$ gives a Riemann sum $\gt U(f,P)-\varepsilon$; so $U(f,P)\lt I+2\varepsilon$. Similarly $L(f,P)\gt I-2\varepsilon$. Thus $U(f,P)-L(f,P)\lt4\varepsilon$; Riemann's criterion gives integrability, and $I$, squeezed between $L$ and $U$ for partitions of arbitrarily small mesh, equals $\int_a^bf$. The final formula is the special case of uniform partitions with right-hand tags. ∎
Proof. (a) A continuous function on the compact interval $[a,b]$ is bounded and uniformly continuous (§3.1, §3.2). Given $\varepsilon\gt0$, choose $\delta\gt0$ with $|f(x)-f(y)|\lt\varepsilon/(b-a)$ whenever $|x-y|\lt\delta$. If $\|P\|\lt\delta$, then on each $I_k$ the extreme value theorem gives points $u,v\in I_k$ with $M_k=f(u)$, $m_k=f(v)$, so $M_k-m_k\lt\varepsilon/(b-a)$ and $U(f,P)-L(f,P)\lt\sum\frac{\varepsilon}{b-a}\Delta x_k=\varepsilon$.
(b) Say $f$ is increasing (otherwise use $-f$). On the uniform partition $P_n$, $M_k=f(x_k)$ and $m_k=f(x_{k-1})$, so the difference telescopes: $$U(f,P_n)-L(f,P_n)=\frac{b-a}{n}\sum_{k=1}^n\big(f(x_k)-f(x_{k-1})\big)=\frac{(b-a)\big(f(b)-f(a)\big)}{n}\to0.$$ (c) Let $|f|\le M$, let $c_1\lt\dots\lt c_r$ be the discontinuities and $\varepsilon\gt0$. Enclose each $c_i$ in an interval $J_i$ of length $\lt\varepsilon/(4Mr)$, the $J_i$ pairwise disjoint. On each of the finitely many closed intervals that remain, $f$ is continuous, so by (a) we can partition it with upper minus lower sum $\lt\varepsilon/(2(r+1))$. Together with the end points of the $J_i$ this gives a partition $P$ of $[a,b]$ with $U-L\lt r\cdot2M\cdot\frac{\varepsilon}{4Mr}+\frac{\varepsilon}{2}=\varepsilon$. ∎
Part (b) is more remarkable than it looks: a monotone function may have infinitely many jumps — countably many, as §3.2 shows — and is still integrable. Part (c) also says that changing a function at finitely many points changes neither its integrability nor its integral (Exercise I.6). To go further we need a way to measure how "small" the set of discontinuities is.
A set $A\subseteq\mathbb{R}$ has (Lebesgue) measure zero (is a null set) if for every $\varepsilon\gt0$ there are countably many open intervals $J_1,J_2,\dots$ with $A\subseteq\bigcup_iJ_i$ and $\sum_i|J_i|\lt\varepsilon$. Every countable set is null (cover the $i$-th point by an interval of length $\varepsilon/2^{i+1}$); a countable union of null sets is null; the Cantor set is an uncountable null set (§4.4). An interval of positive length is not null.
The oscillation of a bounded $f$ at $x$ is $$\omega_f(x)=\lim_{h\to0^+}\Big(\sup_{(x-h,x+h)}f-\inf_{(x-h,x+h)}f\Big)$$ (the intervals intersected with $[a,b]$). $f$ is continuous at $x$ if and only if $\omega_f(x)=0$.
Proof (careful sketch). Write $D_f=\bigcup_{j\ge1}D_j$ with $D_j=\{x:\omega_f(x)\ge1/j\}$. Each $D_j$ is closed (if $\omega_f(x)\lt1/j$, some neighbourhood of $x$ has $\sup-\inf\lt1/j$, and so every point of that neighbourhood has oscillation $\lt1/j$), hence compact.
Integrable $\Rightarrow$ $D_f$ null. It suffices to show each $D_j$ is null, since a countable union of null sets is null. Fix $j$ and $\varepsilon\gt0$, and take $P$ with $U(f,P)-L(f,P)\lt\varepsilon/j$. Let $\mathcal{B}$ be the set of subintervals $I_k$ whose interior meets $D_j$. On such an $I_k$, $M_k-m_k\ge1/j$ (the interior contains a whole neighbourhood of a point of oscillation $\ge1/j$). So $\frac1j\sum_{\mathcal{B}}\Delta x_k\le U-L\lt\varepsilon/j$, i.e. the intervals of $\mathcal{B}$ have total length $\lt\varepsilon$. They cover $D_j$ except possibly the finitely many points of $P$, which we cover by intervals of total length $\lt\varepsilon$. Enlarging slightly to open intervals, $D_j$ is covered by open intervals of total length $\lt3\varepsilon$.
$D_f$ null $\Rightarrow$ integrable. Let $|f|\le M$, $\varepsilon\gt0$ and choose $j$ with $1/j\lt\varepsilon$. The compact set $D_j\subseteq D_f$ is null, so it is covered by countably many open intervals of total length $\lt\varepsilon$, and by compactness (Heine–Borel, §3.1) by finitely many of them, $J_1,\dots,J_r$. On the compact set $K=[a,b]\setminus(J_1\cup\dots\cup J_r)$ every point has $\omega_f\lt1/j$; a compactness argument (cover $K$ by neighbourhoods on which $\sup-\inf\lt1/j$ and take a Lebesgue number) gives a partition of the pieces of $K$ into subintervals on which $M_k-m_k\lt1/j$. Using the end points of the $J_i$ as well, we get a partition $P$ with $$U(f,P)-L(f,P)\le2M\sum_i|J_i|+\frac1j(b-a)\lt\varepsilon\,(2M+b-a).$$ As $\varepsilon$ is arbitrary, Riemann's criterion applies. ∎
Properties of the Integral and the Mean Value Theorems
Let $f,g\in\mathcal{R}[a,b]$ and $\alpha,\beta\in\mathbb{R}$.
- Linearity: $\alpha f+\beta g\in\mathcal{R}[a,b]$ and $\int(\alpha f+\beta g)=\alpha\int f+\beta\int g$.
- Monotonicity: if $f\le g$ then $\int_a^bf\le\int_a^bg$; and $\left|\int_a^bf\right|\le\int_a^b|f|\le M(b-a)$ where $|f|\le M$.
- Additivity: for $a\lt c\lt b$, $f\in\mathcal{R}[a,b]$ iff $f\in\mathcal{R}[a,c]$ and $f\in\mathcal{R}[c,b]$, and $\int_a^bf=\int_a^cf+\int_c^bf$.
- Products and moduli: $fg$, $|f|$, $\max(f,g)$, $\min(f,g)$ are integrable; so is $1/f$ if $|f|\ge c\gt0$.
- Composition: if $m\le f\le M$ and $\varphi:[m,M]\to\mathbb{R}$ is continuous, then $\varphi\circ f\in\mathcal{R}[a,b]$ (e.g. $f^2$, $f^3$, $e^f$, $\sqrt{|f|}$).
- Positivity: if $f$ is continuous, $f\ge0$ and $\int_a^bf=0$, then $f\equiv0$ on $[a,b]$.
Proof. (1) For any $P$, $\sup_{I}(f+g)\le\sup_If+\sup_Ig$ and $\inf_I(f+g)\ge\inf_If+\inf_Ig$, so $$L(f,P)+L(g,P)\le L(f+g,P),$$ $$U(f+g,P)\le U(f,P)+U(g,P).$$ Choosing $P$ (a common refinement) with $U-L\lt\varepsilon/2$ for both $f$ and $g$ gives $U(f+g,P)-L(f+g,P)\lt\varepsilon$ and squeezes $\int(f+g)$ between quantities within $\varepsilon$ of $\int f+\int g$. For $\alpha\ge0$, $U(\alpha f,P)=\alpha U(f,P)$; for $\alpha\lt0$, $U(\alpha f,P)=\alpha L(f,P)$, and the rest follows.
(2) If $f\le g$, every Riemann sum of $f$ is $\le$ the corresponding sum of $g$; pass to the limit (Theorem 4.3). Then $-|f|\le f\le|f|$ gives the modulus inequality once we know $|f|$ is integrable, which is part (4).
(3) Adding the point $c$ to a partition does not increase $U-L$, and a partition of $[a,b]$ containing $c$ is a partition of $[a,c]$ together with one of $[c,b]$; the upper and lower sums split accordingly.
(4), (5) The quickest proof uses Theorem 4.5. A bounded $f$ is integrable iff $D_f$ is null; and $D_{fg}\subseteq D_f\cup D_g$, $D_{|f|}\subseteq D_f$, $D_{\varphi\circ f}\subseteq D_f$ (where $f$ is continuous, so is $\varphi\circ f$), $D_{1/f}=D_f$, while all these functions are bounded. A union of two null sets is null. (A direct proof of (5) with Darboux sums uses the uniform continuity of $\varphi$; for $|f|$ it is just $\big||f(x)|-|f(y)|\big|\le|f(x)-f(y)|$, so $\sup_I|f|-\inf_I|f|\le M_k-m_k$.)
(6) If $f(c)\gt0$ for some $c$, continuity gives an interval $J\ni c$ of length $\ell\gt0$ in $[a,b]$ on which $f\gt f(c)/2$. Then $\int_a^bf\ge\int_Jf\ge\ell f(c)/2\gt0$. ∎
(b) (Weighted form.) If $f$ is continuous on $[a,b]$ and $g\in\mathcal{R}[a,b]$ with $g\ge0$, there is $c\in[a,b]$ with $$\int_a^bf(x)g(x)\,dx=f(c)\int_a^bg(x)\,dx.$$
Proof. (b) Let $m=\min f$ and $M=\max f$ (extreme value theorem). Since $g\ge0$, $mg\le fg\le Mg$, and integrating (Theorem 4.6(2), with $fg$ integrable by 4.6(4)) gives $m\int g\le\int fg\le M\int g$. If $\int g=0$ then $\int fg=0$ and any $c$ works. Otherwise $\int fg/\int g\in[m,M]$, and the intermediate value theorem (§3.1) produces $c$ with $f(c)$ equal to this quotient. (a) is (b) with $g\equiv1$. ∎
The hypothesis $g\ge0$ cannot be dropped: on $[-1,1]$ with $f(x)=g(x)=x$, $\int fg=\tfrac23$ but $\int g=0$. Neither can continuity of $f$: a step function need not take its average value.
The Fundamental Theorem of Calculus
For $f\in\mathcal{R}[a,b]$ the indefinite integral $F(x)=\int_a^xf(t)\,dt$ ($a\le x\le b$) is defined by Theorem 4.6(3). The two halves of the fundamental theorem say that integrating and differentiating undo each other — under hypotheses that matter.
- $|F(x)-F(y)|\le M|x-y|$ for all $x,y\in[a,b]$: $F$ is Lipschitz, hence uniformly continuous;
- if $f$ is continuous at $c\in[a,b]$, then $F$ is differentiable at $c$ and $F'(c)=f(c)$ (one-sided at $c=a,b$);
- more generally, if $f(c^+)=\lim_{t\to c^+}f(t)$ exists, the right derivative of $F$ at $c$ equals $f(c^+)$; similarly on the left.
Proof. (1) For $x\lt y$, $F(y)-F(x)=\int_x^yf$ by additivity, and $\left|\int_x^yf\right|\le M(y-x)$.
(2) and (3) For $h\gt0$ small, $$\frac{F(c+h)-F(c)}{h}-f(c^+)=\frac1h\int_c^{c+h}\big(f(t)-f(c^+)\big)\,dt.$$ Given $\varepsilon\gt0$ choose $\delta$ with $|f(t)-f(c^+)|\lt\varepsilon$ for $c\lt t\lt c+\delta$. For $0\lt h\lt\delta$ the right side has modulus $\le\frac1h\cdot\varepsilon h=\varepsilon$ (the value of the integrand at the single point $t=c$ does not matter, Theorem 4.4(c)). So the right derivative is $f(c^+)$; the left derivative is $f(c^-)$ in the same way. If $f$ is continuous at $c$, both equal $f(c)$. ∎
Proof. Let $P$ be any partition. By the mean value theorem (§3.3) on each $I_k$ there is $c_k\in(x_{k-1},x_k)$ with $F(x_k)-F(x_{k-1})=F'(c_k)\Delta x_k$. Summing, the left side telescopes: $$F(b)-F(a)=\sum_{k=1}^nf(c_k)\,\Delta x_k,$$ a Riemann sum, which lies between $L(f,P)$ and $U(f,P)$. As this holds for every $P$, $\underline{\int}f\le F(b)-F(a)\le\overline{\int}f$, and the two outer quantities are both $\int_a^bf$. For the extension, apply the result on each closed piece between exceptional points (the mean value theorem needs only continuity on $[x_{k-1},x_k]$ and differentiability inside). ∎
Leibniz's rule. If $f$ is continuous on an interval $J$ and $u,v$ are differentiable functions with values in $J$, then by Theorem 4.8 and the chain rule, $$\frac{d}{dx}\int_{u(x)}^{v(x)}f(t)\,dt=f\big(v(x)\big)v'(x)-f\big(u(x)\big)u'(x).$$ (Write the integral as $G(v(x))-G(u(x))$ with $G(y)=\int_{y_0}^yf$, $y_0\in J$.)
(b) If $\varphi:[\alpha,\beta]\to\mathbb{R}$ has a continuous derivative and $f$ is continuous on an interval containing $\varphi([\alpha,\beta])$, then $$\int_\alpha^\beta f\big(\varphi(t)\big)\varphi'(t)\,dt=\int_{\varphi(\alpha)}^{\varphi(\beta)}f(x)\,dx.$$
Proof. (a) $u,v$ are continuous, hence integrable, so $(uv)'=u'v+uv'$ is integrable (Theorem 4.6). Theorem 4.9 applied to $uv$ gives $\int(u'v+uv')=[uv]_a^b$. (b) Let $G(y)=\int_{\varphi(\alpha)}^yf$; by Theorem 4.8, $G'=f$, so $(G\circ\varphi)'(t)=f(\varphi(t))\varphi'(t)$, a continuous function. Theorem 4.9 gives $\int_\alpha^\beta f(\varphi(t))\varphi'(t)\,dt=G(\varphi(\beta))-G(\varphi(\alpha))$, which is $\int_{\varphi(\alpha)}^{\varphi(\beta)}f$. Note that $\varphi$ need not be monotone or injective. ∎
- Is it bounded? If not, it is not Riemann integrable (proper sense) — stop, or treat it as an improper integral (§4.2).
- Where is it discontinuous? Find $D_f$. Finite or countable (jumps of a monotone function, the rationals, $\{1/n\}\cup\{0\}$) $\Rightarrow$ integrable. A Cantor-type closed set: decide whether its measure is $0$. An interval, or "every irrational", or a set of positive measure $\Rightarrow$ not integrable.
- Value of the integral. If $f=g$ except on a finite set, $\int f=\int g$. If $f\ge0$ vanishes on a dense set (e.g. at all irrationals) and is integrable, every lower sum is $0$, so $\int f=0$.
- Limits of sums. Look for $\tfrac1n\sum f(k/n)$ — after taking logarithms for products, and after a change of index for ranges like $k=n+1,\dots,2n$ (giving $\int_1^2$).
- $F(x)=\int_a^xf$. Always Lipschitz; $F'(c)=f(c)$ at continuity points; one-sided derivatives $f(c^\pm)$ at jumps; nothing can be concluded about $f$ from the differentiability of $F$ at a single point.
- Bounded but not integrable. The Dirichlet function $\mathbf 1_{\mathbb{Q}}$: $U=1$, $L=0$ for every partition.
- $|f|$ or $f^2$ integrable does not make $f$ integrable. $f=1$ on $\mathbb{Q}$, $-1$ off $\mathbb{Q}$: $|f|=f^2\equiv1$, but $D_f=[a,b]$. (The converse is true: $f$ integrable $\Rightarrow|f|,f^2$ integrable.)
- Composition of two integrable functions. With Thomae's $T$ and $\varphi(y)=0$ for $y=0$, $\varphi(y)=1$ otherwise (integrable, one discontinuity), $\varphi\circ T=\mathbf 1_{\mathbb{Q}}$ on $[0,1]$. Theorem 4.6(5) needs the outer function continuous. Even "integrable $\circ$ continuous" can fail: a homeomorphism $g$ of $[0,1]$ carrying a fat Cantor set $K$ onto the Cantor set $C$ gives $\mathbf 1_C\circ g=\mathbf 1_K$.
- Countable is sufficient, not necessary. $\mathbf 1_C$ ($C$ the Cantor set) is integrable with uncountably many discontinuities; $\mathbf 1_K$ for a fat Cantor set is not. Measure, not cardinality, decides.
- Changing values. Changing $f$ at finitely many points changes nothing; changing it at countably many points can destroy integrability ($0\mapsto\mathbf 1_{\mathbb{Q}}$).
- Limits of integrable functions. $f_n=\mathbf 1_{\{r_1,\dots,r_n\}}$ (an enumeration of $\mathbb{Q}\cap[0,1]$) are integrable with $\int f_n=0$, and $f_n\to\mathbf 1_{\mathbb{Q}}$ pointwise — not integrable. Uniform limits behave well (Exercise IV.4, Chapter 5).
- One sequence of Riemann sums is not enough. $\frac1n\sum_{k=1}^n\mathbf 1_{\mathbb{Q}}(k/n)=1$ for all $n$, yet $\mathbf 1_{\mathbb{Q}}$ is not integrable. Theorem 4.3 requires convergence for all tags; but once integrability is known, any sequence of tagged partitions with mesh $\to0$ works.
- An integrable function need not have an antiderivative. $\operatorname{sign}x$ on $[-1,1]$ is integrable, but no $F$ has $F'=\operatorname{sign}$ everywhere: derivatives have the intermediate value property (Darboux, §3.3).
- A derivative need not be integrable. $F(x)=x^2\sin(1/x^2)$, $F(0)=0$, is differentiable on $[0,1]$ but $F'$ is unbounded near $0$. (Volterra even built an $F$ with $F'$ bounded but not Riemann integrable.) Theorem 4.9 must assume $F'\in\mathcal{R}$.
- $F$ differentiable at $c$ says nothing about $f$ at $c$. $f=\mathbf 1_{\{0\}}$: $F\equiv0$, $F'(0)=0\ne f(0)$.
- Positivity needs continuity. $f\ge0$, $\int f=0$ forces $f\equiv0$ only for continuous $f$: Thomae's function is positive at every rational and has integral $0$.
- Mean value theorem needs $g\ge0$. $\int_{-1}^1x\cdot x\,dx=\tfrac23$ while $\int_{-1}^1x\,dx=0$, so no $c$ satisfies $\int fg=f(c)\int g$.
Exercises
- 1 For $f(x)=x^3$ on $[0,2]$ and the uniform partition $P_n$, compute $U(f,P_n)$, $L(f,P_n)$ and $U-L$, and find $\int_0^2x^3\,dx$.
- 2 For $f(x)=1/x$ on $[1,2]$ use the geometric partition $x_k=2^{k/n}$ ($0\le k\le n$). Show $U(f,P)=n(2^{1/n}-1)$ and $L(f,P)=n(1-2^{-1/n})$, and deduce $\int_1^2dx/x=\log2$.
- 3 Let $f(x)=0$ for $x\ne\tfrac12$ and $f(\tfrac12)=5$, on $[0,1]$. Prove from Riemann's criterion that $f$ is integrable with $\int_0^1f=0$.
- 4 Let $f(x)=1$ for rational $x$ and $f(x)=x$ for irrational $x$, on $[0,1]$. Find $\overline{\int_0^1}f$ and $\underline{\int_0^1}f$.
- 5 Prove: if $f\in\mathcal{R}[a,b]$ then $f\in\mathcal{R}[c,d]$ for every $[c,d]\subseteq[a,b]$.
- 6 Prove that if $f\in\mathcal{R}[a,b]$ and $g=f$ except at one point, then $g\in\mathcal{R}[a,b]$ and $\int g=\int f$. Extend to finitely many points.
- 7 Without Lebesgue's criterion, prove that $f(x)=\sin(1/x)$ ($x\ne0$), $f(0)=0$, is integrable on $[0,1]$.
- 8 Prove that a bounded $f$ is integrable on $[a,b]$ iff for every $\varepsilon\gt0$ there are step functions $s\le f\le t$ with $\int_a^b(t-s)\lt\varepsilon$.
- 1 Which are Riemann integrable on $[0,1]$? (a) $f(x)=x\lfloor1/x\rfloor$, $f(0)=0$; (b) $f(x)=\operatorname{sign}\big(\sin(1/x)\big)$, $f(0)=0$; (c) $f(x)=x(1-x)\mathbf 1_{\mathbb{Q}}(x)$; (d) $f(x)=\lfloor1/x\rfloor$, $f(0)=0$.
- 2 Compute $\displaystyle\int_0^2\lfloor x^2\rfloor\,dx$.
- 3 For Thomae's function $T$ and the uniform partition of $[0,1]$ into $4$ subintervals, compute $U(T,P_4)$.
- 4 Prove directly (with Darboux sums) that if $f\in\mathcal{R}[a,b]$ and $f\ge c\gt0$, then $1/f\in\mathcal{R}[a,b]$.
- 5 Prove the Cauchy–Schwarz inequality $\left(\int_a^bfg\right)^2\le\int_a^bf^2\int_a^bg^2$ for $f,g\in\mathcal{R}[a,b]$.
- 6 Let $f$ be continuous on $[a,b]$ with $\int_a^bf(x)g(x)\,dx=0$ for every continuous $g$. Prove $f\equiv0$.
- 7 Prove that if $f\in\mathcal{R}[a,b]$, the points of continuity of $f$ are dense in $[a,b]$.
- 8 Find a point $c$ given by the weighted mean value theorem for $\int_0^1x^2\cdot e^x\,dx$ with $f(x)=x^2$, $g(x)=e^x$ (give $c$ to three decimals).
- 1 Evaluate $\displaystyle\lim_{n\to\infty}\sum_{k=1}^n\frac1{\sqrt{n^2+kn}}$.
- 2 Evaluate $\displaystyle\lim_{n\to\infty}\sum_{k=1}^n\frac{k^2}{n^3+k^3}$.
- 3 Evaluate $\displaystyle\lim_{n\to\infty}\frac1{n^2}\sum_{k=1}^nk\sin\frac{k\pi}n$.
- 4 Evaluate $\displaystyle\lim_{n\to\infty}\frac1n\sum_{k=1}^ne^{k/n}$, and compare the sum for $n=10$ with the limit.
- 5 Find $F'(x)$ for $F(x)=\displaystyle\int_0^{x^2}\sqrt{1+t^3}\,dt$.
- 6 Find $G'(x)$ for $G(x)=\displaystyle\int_{\sin x}^{\cos x}e^{t^2}\,dt$.
- 7 Evaluate $\displaystyle\lim_{x\to0^+}\frac1{x^3}\int_0^{x^2}\sin\sqrt t\,dt$.
- 8 Find all continuous $f:\mathbb{R}\to\mathbb{R}$ with $f(x)=1+\displaystyle\int_0^xt\,f(t)\,dt$ for all $x$.
- 9 Let $f(t)=t$ for $0\le t\lt1$ and $f(t)=3$ for $1\le t\le2$. Find $F(x)=\int_0^xf$ and the points where $F$ is not differentiable.
- 1 (Riemann–Lebesgue lemma) Prove that $\displaystyle\lim_{n\to\infty}\int_a^bf(x)\sin(nx)\,dx=0$ for every $f\in\mathcal{R}[a,b]$. (First do it for step functions.)
- 2 Let $f$ be continuous on $[0,\pi]$ with $\int_0^\pi f(x)\sin x\,dx=\int_0^\pi f(x)\cos x\,dx=0$. Prove that $f$ has at least two zeros in $(0,\pi)$.
- 3 Let $f\in\mathcal{R}[0,1]$ satisfy $\int_0^xf=\int_x^1f$ for every $x\in[0,1]$. Prove that $f(c)=0$ at every point $c$ of continuity of $f$.
- 4 Prove that a uniform limit of Riemann integrable functions on $[a,b]$ is Riemann integrable, and that the integrals converge.
- 5 Let $f$ be continuous on $[0,\infty)$ with $f(x)\to L$ as $x\to\infty$. Prove $\frac1x\int_0^xf(t)\,dt\to L$.
- 6 Let $f$ have a continuous derivative on $[0,1]$. Prove $$n\Big(\frac1n\sum_{k=1}^nf\big(\tfrac kn\big)-\int_0^1f\Big)\to\frac{f(1)-f(0)}2.$$
- 7 Let $f$ be differentiable on $[0,1]$ with $f'\in\mathcal{R}[0,1]$ and $f(0)=0$. Prove $\displaystyle\int_0^1f^2\le\frac12\int_0^1(f')^2$.
E1. CSIR Part B GATE MA The value of $\displaystyle\lim_{n\to\infty}\sum_{k=1}^n\frac{n+k}{n^2+k^2}$ is (A) $\dfrac\pi4$; (B) $\dfrac\pi4+\dfrac{\log2}2$; (C) $\log2$; (D) $\dfrac\pi2-\log2$.
E2. CSIR Part B Let $F(x)=\int_0^x\lfloor t\rfloor\,dt$ for $x\in[0,3]$. Which statement is true? (A) $F$ is differentiable on $(0,3)$; (B) $F$ is continuous on $[0,3]$ and not differentiable exactly at $x=1$ and $x=2$; (C) $F$ is discontinuous at $1$ and $2$; (D) $F'(x)=\lfloor x\rfloor$ for every $x\in(0,3)$.
E3. CSIR Part B Define $f:[0,1]\to\mathbb{R}$ by $f(x)=1$ if $x=\tfrac1n$ for some $n\in\mathbb{N}$, and $f(x)=0$ otherwise. Then (A) $f$ is not Riemann integrable, since it has infinitely many discontinuities; (B) $f$ is Riemann integrable and $\int_0^1f=0$; (C) $f$ is Riemann integrable and $\int_0^1f=1$; (D) the upper integral of $f$ is $1$.
E4. CSIR Part C Let $f:[0,1]\to\mathbb{R}$ be bounded. Which of the following statements are true? (One or more options may be correct.) (A) If $f^2$ is Riemann integrable, then $f$ is Riemann integrable. (B) If $f$ is Riemann integrable, then $f^3$ is Riemann integrable. (C) If $f$ is continuous except at countably many points, then $f$ is Riemann integrable. (D) If $f$ is Riemann integrable and $g:[0,1]\to\mathbb{R}$ is bounded with $g=f$ except at countably many points, then $g$ is Riemann integrable.
E5. CSIR Part C Let $f:[-1,1]\to\mathbb{R}$ be Riemann integrable and $F(x)=\int_0^xf(t)\,dt$. Which of the following are true? (A) $F$ is uniformly continuous on $[-1,1]$. (B) If $f$ is continuous at $0$, then $F'(0)=f(0)$. (C) If $F$ is differentiable at $0$, then $f$ is continuous at $0$. (D) If $\lim_{t\to0^-}f(t)$ and $\lim_{t\to0^+}f(t)$ exist and are different, then $F$ is not differentiable at $0$.
E6. CSIR Part C Let $f:[0,1]\to\mathbb{R}$ be continuous. Which of the following hold for every such $f$? (A) $\displaystyle\lim_{n\to\infty}\int_0^1f(x^n)\,dx=f(0)$. (B) $\displaystyle\lim_{n\to\infty}n\int_0^1x^nf(x)\,dx=f(1)$. (C) $\displaystyle\lim_{n\to\infty}\int_0^1nf(x)e^{-nx}\,dx=f(0)$. (D) $\displaystyle\lim_{n\to\infty}\int_0^1f(x)\,|\sin nx|\,dx=0$.
Suggested plan — 3 lectures × 60 min.
- Darboux sums. Definitions 4.1–4.3, the refinement lemma with a picture of one added point, Riemann's criterion (Theorem 4.2). Example 1 on the board, then Interactive 1 with $x^2$ and the step function; switch to the Dirichlet function and let the class explain why the amber never shrinks (Example 2).
- Which functions are integrable? Theorem 4.3 (state, prove one direction), Theorem 4.4 with the telescoping proof for monotone functions, measure zero and Lebesgue's criterion (state, sketch the "$\Leftarrow$" direction with the oscillation sets $D_j$). Example 3 (Thomae) with Interactive 1, Example 5, and the pitfalls on compositions and $|f|$.
- Properties and FTC. Theorems 4.6–4.10, Example 4 (limits of sums — have students find the integral before you reveal it), Example 7 and Interactive 2 (cycle through all five integrands; stop at the one-point function). Finish with Exam-corner E2 and E5 live.
Discussion questions. (1) Thomae's function and the Dirichlet function both "see" the rationals. Why is one integrable and the other not? (2) If $F$ is differentiable everywhere, is $\int_a^bF'=F(b)-F(a)$ automatic? What extra hypothesis does Theorem 4.9 need, and why? (3) Can a function be integrable on $[0,1]$ and discontinuous at uncountably many points? Can its discontinuities be dense? Both?
Misconceptions to watch for. "Infinitely many discontinuities ⇒ not integrable"; "uncountably many discontinuities ⇒ not integrable" (the Cantor set); "$|f|$ integrable ⇒ $f$ integrable"; computing a limit of a sum by one convenient choice of tags for a function not yet known to be integrable; "$F'=f$ everywhere" for any integrable $f$; forgetting boundedness.
Suggested homework. I(1, 2, 4, 7), II(1, 2, 3, 5, 7), III(1, 2, 3, 6, 8, 9), IV(1, 3, 5); Exam corner E1, E3, E4, E6.
Research question. How fast do Riemann sums converge, and how does the answer depend on the choice of tags and on the smoothness of the integrand?
Why it matters. In clinical pharmacology the "area under the curve" (AUC) of a drug's blood concentration measures total exposure and is used worldwide to compare generic and brand-name medicines; regulators accept AUC computed by the trapezoidal rule from a handful of blood samples. Environmental engineers estimate the total pollutant load of a river from sampled flow rates in the same way. Knowing the error of such sums tells you how many samples are needed.
What to do.
- For $f(x)=e^x$ on $[0,1]$ compute the left sum $L_n$, midpoint sum $M_n$, trapezoidal sum $T_n=\tfrac12(L_n+R_n)$ and Simpson's value $S_n=\tfrac13(T_n+2M_n)$ for $n=10,20,40,80$.
- Tabulate the errors and estimate the exponent $p$ in error $\approx Cn^{-p}$ from successive ratios. Explain the result for $L_n$ with Exercise IV.6.
- Repeat for $f(x)=\sqrt x$, whose derivative is unbounded at $0$. Which rules lose their speed, and why?
- Model an AUC study: take $c(t)=10\,(e^{-0.2t}-e^{-1.5t})$ mg/L for $0\le t\le24$ h, "sample" it at $8$ irregular times, and compare the trapezoidal AUC with the exact integral.
Tools: Python with NumPy (or a spreadsheet); Interactive 1 for intuition about tags.
Hand in: A one-page report with the error table, the fitted exponents, and a recommendation on sampling times for the AUC model.
Going further. Read about the Euler–Maclaurin formula, which explains the trapezoidal exponent and why the trapezoidal rule is spectacularly accurate for smooth periodic functions; then try $\int_0^{2\pi}e^{\cos x}dx$.
🧪 Self-check quiz
Q1. Which function is not Riemann integrable on $[0,1]$?
Q2. $\displaystyle\lim_{n\to\infty}\sum_{k=1}^n\frac1{n+k}$ equals
Q3. If $F(x)=\int_0^{x^3}\cos(t^2)\,dt$, then $F'(x)=$
Q4. Riemann's criterion says that a bounded $f$ is integrable on $[a,b]$ if and only if
Q5. Let $f\in\mathcal{R}[a,b]$ and $F(x)=\int_a^xf$. Which statement is always true?